Overview. Background / Context. CSC 580 Cryptography and Computer Security. March 21, 2017

Transcription

1 CSC 580 Cryptography and Computer Security Math for Public Key Crypto, RSA, and Diffie-Hellman (Sections , 2.8, 9.2, ) March 21, 2017 Overview Today: Math needed for basic public-key crypto algorithms RSA and Diffie-Hellman Next: Read Chapter 11 (skip SHA-512 logic and SHA3 iteration function) Project phase 3 due in one week (March 28) - finish it! Background / Context Recall example trapdoor function from last time: Given a number n, how many positive integers divide evenly into n? If you know the prime factorization of n, this is easy. If you don t know the factorization, don t know efficient solution How does this fit into the public key crypto model? Pick two large (e.g., 1024-bit) prime numbers p and q Compute the product n = p * q Public key is n (hard to find p and q!), private is the pair (p,q) Questions: How do we pick (or detect) large prime numbers? How do we use this trapdoor knowledge to encrypt?

2 Prime Numbers A prime number is a number p for which its only positive divisors are 1 and p Question: How common are prime numbers? The Prime Number Theorem states that there are approximately n / ln n prime numbers less than n. Picking a random b-bit number, probability that it is prime is approximately 1/ln(2 b ) = (1/ln 2)*(1/b) 1.44 * (1/b) For 1024-bit numbers this is about 1/710 Pick random 1024-bit numbers until one is prime takes on average 710 trials This is efficient - if we can tell when a number is prime! Primality Testing Problem: Given a number n, is it prime? Basic algorithm: Try dividing all numbers 2,..,sqrt(n) into n Question: How long does this take if n is 1024 bits? Fermat s Little Theorem To do better, we need to understand some properties of prime numbers, such as Fermat s Little Theorem: If p is prime and a is a positive integer not divisible by p, then a p-1 1 (mod p). Proof is on page 46 of the textbook (not difficult!).

3 Fermat s Little Theorem - cont d Explore this formula for different values of n and random a s: a (n = 221) (n = 331) (n = 441) (n = 541) Question 1: What conclusion can be drawn about the primality of 221? Question 2: What conclusion can be drawn about the primality of 331? Primality Testing - First Attempt Tempting (but incorrect) primality testing algorithm for n: Pick random a {2,..., n-2} if 1 then return not prime else return probably prime Why doesn t this work? Primality Testing - First Attempt Tempting (but incorrect) primality testing algorithm for n: Pick random a {2,..., n-2} if 1 then return not prime else return probably prime Why doesn t this work? Carmichael numbers. a (n = 2465) Example: 2465 is obviously not prime, but Note: Not just for these a s, but = 1 for all a s that are relatively prime to n

4 Primality Testing - Miller-Rabin The previous idea is good, with some modifications (Note: This corrects a couple of typos in the textbook): MILLER-RABIN-TEST(n) // Assume n is odd Find k>0 and q odd such that n-1 = 2 k q Pick random a {2,..., n-2} x = a q mod n if x = 1 or x = n-1 then return possible prime for j = 1 to k-1 do x = x 2 mod n if x = n-1 then return possible prime return composite If n is prime, always returns possible prime If n is composite, says possible prime with probability < ¼ Idea: Run 50 times, and accept as prime iff all say possible prime Question: What is the error probability? Euler s Totient Function and Theorem Euler s totient function: (n) = number of integers from 1..n-1 that are relatively prime to n. If s(n) is count of 1..n-1 that share a factor with n, (n) = n s(n) s(n) was our trapdoor function example (n) easy to compute if factorization of n known Don t know how to efficiently compute otherwise If n is product of two primes, n=p*q, then s(n)=(p-1)+(q-1)=p+q-2 So (p*q) = p*q (p+q-2) = p*q - p - q + 1 = (p-1)*(q-1) Euler generalized Fermat s Little Theorem to composite moduli: Euler s Theorem: For every a and n that are relatively prime (i.e., gcd(a,n)=1), a (n) 1 (mod n). Question: How does this simplify if n is prime? RSA Algorithm Key Generation: Pick two large primes p and q Calculate n=p*q and (n)=(p-1)*(q-1) Pick a random e such that gcd(e, (n)) Compute d = e -1 (mod (n)) [Use extended GCD algorithm!] Public key is PU=(n,e) ; Private key is PR=(n,d) Encryption of message M {0,..,n-1}: E(PU,M) = M e mod n Decryption of ciphertext C {0,..,n-1}: D(PR,C) = C d mod n

5 RSA Algorithm Key Generation: Pick two large primes p and q Calculate n=p*q and (n)=(p-1)*(q-1) Pick a random e such that gcd(e, (n)) Compute d = e -1 (mod (n)) [Use extended GCD algorithm!] Public key is PU=(n,e) ; Private key is PR=(n,d) Correctness - easy when gcd(m,n)=1: Encryption of message M {0,..,n-1}: E(PU,M) = M e D(PR,E(PU,M)) = (M e ) d mod n mod n = M ed mod n = M k (n)+1 mod n Decryption of ciphertext C {0,..,n-1}: = (M (n) ) k M mod n = M D(PR,C) = C d mod n Also works when gcd(m,n) 1, but slightly harder to show... RSA Example Simple example: p = 73, q = 89 n = p*q = 73*89 = 6497 (n) = (p-1)*(q-1) = 72*88 = 6336 e = 5 d = 5069 [ Note: 5*5069 = 25,345 = 4* ] Encrypting message M=1234: mod 6497 = 1881 Decrypting: mod 6497 = 1234 Note: If time allows in class, more examples using Python! The Discrete Log Problem For every prime number p, there exists a primitive root (or generator ) g such that g 1, g 2, g 3, g 4,, g p-2, g p-1 (all taken mod p) are all distinct values (so a permutation of 1, 2, 3,..., p-1). Example: 3 is a primitive root of 17, with powers: i i mod f g,p (i) = g i mod p is a bijective mapping on {1,.., p-1} f g,p (i) is easy to compute (modular powering algorithm) g and p are global public parameters Inverse, written dlog g,p (x) = f g,p -1 (x), is believed to be difficult to compute

6 Diffie-Hellman Key Exchange Assume g and p are known, public parameters Alice a random value from {1,, p-1} A g a mod p Bob b random value from {1,, p-1} B g b mod p Send A to Bob Send B to Alice S a B a mod p S b A b mod p In the end, Alice s secret (S a ) is the same as Bob s secret (S b ): S a = B a = g ba = g ab = A b = S b Eavesdropper knows A and B, but to get a or b requires solving the discrete logarithm problem! Abstracting the Problem There are many sets over which we can define powering. Example: Can look at powers of n n matrices (A 2, A 3, etc.) Any finite set S with an element g such that f g : S S is a bijection (where f g (x) = g x for all x S) is called a cyclic group Very cool math here - see Chapter 5 for more info (optional) If f g is easy to compute and f g -1 is difficult, then can do Diffie-Hellman Elliptic Curves are a mathematical object with this property In fact: f g -1 seems to be harder to compute for Elliptic Curves than Z p Consequence: Elliptic Curves can use shorter numbers/keys than standard Diffie-Hellman - so faster and less communication required! Revisiting Key Sizes From NIST publication a Issue: PK algorithms based on mathematical relationships, and can be broken with algorithms that are faster than brute force. We spent time getting a feel for how big symmetric cipher\ keys needed to be How big do keys in a public key system need to be? From NIST pub a: